Most static alerts are false positives caused by Markdown backticks and payment-discovery language in documentation. Two confirmed high-severity findings describe exporting an agent private key in a JSON package, which can expose a bearer capability and permit unauthorized delegated spending. No prompt-injection language, data-exfiltration intent, or unscoped command execution was found.
- The page generates a real agent EOA for the demo and exports a JSON package containing the agent p
The skill directs the demo to generate a real agent EOA and export its private key in a JSON package. Exported private keys are bearer secrets and can enable unauthorized spending if exposed.
- include private key in exported JSON or omit it;
The implementation menu presents including a private key in exported JSON as an option. This normalizes an insecure secret-export path that can expose the delegated wallet capability.
Most command and reconnaissance alerts are false positives caused by Markdown code spans, TypeScript examples, and payment-demo language. The remaining high-severity issue is secret handling because the skill permits exporting a generated agent private key in a demo package.
- The page generates a real agent EOA for the demo and exports a JSON package containing the agent p
The skill instructs a demo page to export a JSON package containing an agent private key. That artifact is a bearer secret that could spend delegated authority if exposed.
- include private key in exported JSON or omit it;
The skill presents including a private key in exported JSON as an implementation choice. Even for demos, this creates a secret-handling risk if the artifact is shared or reused.